Google Tensor SDK beta adds LiteRT deployment for Pixel 10 on-device AI

ByManjit Singh

Google has moved Tensor ML SDK from experimental access to beta, with LiteRT now integrated for a more unified on-device workflow on Pixel 10 devices.

For builders, the key change is that Tensor ML SDK now supports converting, compiling, deploying, and running PyTorch or TFLite models on Google Tensor’s TPU through LiteRT. Google also says the runtime can fall back to CPU or GPU when TPU execution is unavailable.

The beta launch also includes a model garden with 100+ models, including Gemma 3 1B and precompiled models available through the LiteRT Hugging Face community.

Why it matters

This lowers friction for shipping private, low-latency AI features on-device. If you build mobile or edge AI products, it’s worth checking whether your pipeline can target LiteRT and whether your app can tolerate fallback behavior across devices and execution paths.

Builder takeaway

Evaluate Tensor SDK + LiteRT if you want to:

  • run inference on-device instead of in the cloud
  • reduce latency for interactive AI features
  • keep certain workloads local for privacy
  • reuse PyTorch or TFLite models in a Pixel 10 deployment flow

Google says the beta supports Pixel 10, Pixel 10 Pro, Pixel 10 Pro XL, and Pixel 10 Pro Fold.

Similar Posts